Overview
- OpenAI rolled out the new image model on Wednesday in ChatGPT, Codex and a gpt-image-2 API.
- A new Thinking mode plans image layouts, can draw on recent web information and produces up to eight consistent images in one run, with access limited to Plus, Pro and Business tiers.
- Text inside images is markedly clearer, including small labels, UI icons and non‑Latin scripts, which opens the door to menus, infographics and other production assets.
- Outputs look more natural thanks to better light, textures and fine details, and the model supports aspect ratios from 3:1 to 1:3 while keeping style and characters consistent across related images.
- API requests above 2K resolution remain in beta and error‑prone, and pricing varies by quality and size as OpenAI shares few technical details beyond noting reasoning and self‑checks.